AMD Strix Halo Toolboxes — Setup Notes

Tested on Ubuntu 24.04 LTS.

System update

Run as root or prefix with sudo:

sudo apt update && sudo apt upgrade -y
sudo reboot

Install AMD GPU packages (example: amdgpu/ROCm)

Download the installer (example version used here: 7.0.2):

wget https://repo.radeon.com/amdgpu-install/7.0.2/ubuntu/noble/amdgpu-install_7.0.2.70002-1_all.deb
sudo apt install ./amdgpu-install_7.0.2.70002-1_all.deb
sudo apt update
sudo apt install python3-setuptools python3-wheel

Add current user to render/video groups:

sudo usermod -a -G render,video $LOGNAME

Install ROCm:

sudo apt install rocm

Kernel extras and DKMS:

sudo apt install "linux-headers-$(uname -r)" "linux-modules-extra-$(uname -r)"
sudo apt install amdgpu-dkms
sudo reboot

Library path and ldconfig

Add ROCm libs to the loader path and refresh:

sudo tee /etc/ld.so.conf.d/rocm.conf <<'EOF'
/opt/rocm/lib
/opt/rocm/lib64
EOF
sudo ldconfig

You can also set LD_LIBRARY_PATH for the current session (or add to your shell profile):

export LD_LIBRARY_PATH=/opt/rocm-7.0.2/lib:$LD_LIBRARY_PATH

Verify installation:

apt list --installed | grep -E 'rocm|amdgpu'
rocminfo | grep -i 'Marketing Name:'

Kernel / AMD TTM tuning

Edit GRUB:

sudo nano /etc/default/grub
# add or update GRUB_CMDLINE_LINUX_DEFAULT with required options, for example:
GRUB_CMDLINE_LINUX_DEFAULT="quiet amdttm.pages_limit=33554432 amdgpu.gttsize=131072 amd_iommu=off"
sudo update-grub
sudo reboot
# verify:
sudo dmesg | grep -i 'amdgpu.*memory'

Toolbox / Container workflow

Install Podman Toolbox if desired:

sudo apt-get install podman-toolbox

Clone the repo:

git clone https://github.com/kyuz0/amd-strix-halo-toolboxes.git
cd amd-strix-halo-toolboxes/toolboxes/

Create a toolbox container (example):

sudo toolbox create llama-rocm-7rc-rocwmma \
  --image docker.io/kyuz0/amd-strix-halo-toolboxes:rocm-7rc-rocwmma \
  -- --device /dev/dri --device /dev/kfd \
     --group-add video --group-add render --group-add sudo \
     --security-opt seccomp=unconfined

Enter the toolbox:

toolbox enter llama-rocm-7rc-rocwmma

Inside the container you can list devices:

llama-cli --list-devices

Run a server example (adjust model name and flags as needed):

llama-server --no-mmap -ngl 999 -fa on --host 0.0.0.0 -hf unsloth/Qwen3-235B-A22B-Instruct-2507-GGUF:Q3_K_XL

Notes

Always check official AMD docs and AMD-Strix-Halo Repo for changes